Few AI Startups Release Revenue Figures, Much Less Income: Can They Be Profitable?

Published in

The Startup

6 min readDec 22, 2020

Only four Western and two Chinese AI companies report income, and all have big losses. CrowdStrike and c3.ai both did IPOs and had losses equal to 30% and 40% of revenues respectively in 2019, and 13% and 40% respectively in 2020. Nest’s losses were 85% of revenues in 2017[1] and DeepMind’s losses were four and 1.7 times its revenues in 2018[2] and 2019 respectively[3] causing Google to write off $1.3 Billion in debts. For China, Megvii’s cumulative losses had reached $1.4 billion by the first half of 2019, or 2.5 times cumulative revenues[4]. CloudMinds reported a net loss of US$97.48 million in the six months of 2019 or 2/3 of its reported revenues[5]. Furthermore, the big funding received by many Chinese (e.g., CloudWalk, Yitu, SenseTime) and American startups (described below) suggest most if not all AI startups are unprofitable.

Of 40 American startups I analyzed for an IEEE Spectrum article published in February 2020, there is also little evidence of profitability or even of growth. One went bankrupt, another did an IPO and has big losses, two have been acquired (FlatIron by Roche, Cylance by Blackberry), a Chinese startup closed its American headquarters, and one has reoriented towards bio-manufacturing (Zymergen). Of the remaining 34 (See Table), only one reports rough figures for revenues (UI Path), the rest don’t, and 22 of them have received new funding in the last two 18 months, since I did the analysis for the IEEE Spectrum article. The article argued that few of them offered products and services that directly impact on productivity, and this article suggests most if not all of them are a long way from profitability, a conclusion consist with other startups, particularly Unicorn startups[6].

Venture Capital Funding ($M) and Industry for Top Funded Startups

The large number of startups receiving recent funding is a double-edged sword. On one hand, recent funding means VCs are optimistic about the startups, perhaps because the startups are experiencing growth and/or because the cost of capital is so low. On the other hand, the additional funding suggests they have big losses and thus are far from profitability. After all, the money is needed for salaries and not for building warehouses and other parts of a logistic system, which is what Amazon was doing when it incurred losses (profits in year 10). Five startups have received $1 Billion or more in funding. Furthermore, seven have received funding since the U.S. federal government introduced stimulus packages in Spring 2020, thus making it even easier to raise funds, perhaps unwisely; some of these startups have received funding in multiple rounds in the last two years.

The largest startups are the most controversial, and not just for their large losses. For instance, DeepMind (and other parts of Google) has generated controversary through claims it reduced energy usage at a Google data center, it could do the same for the UK economy (perhaps in combination with Nest), its AI program had outperformed humans in diagnosing breast cancer [7], and it solved the protein folding problem. No independent verifications have been reported for these cost reductions and a 2019 Economist article commented on the reduction in data center costs, saying “some insiders say such boasts are overblown[8]. A group led by Benjamin Haibe-Kains, a computational genomics researcher, criticized the breast cancer paper, arguing that the “lack of details of the methods and algorithm code undermines its scientific value [9].” Others called “laughable” the claims that the protein folding problem had been solved and said that this research will “never live up to promise that’s been made[10].”

China’s largest AI companies are equally controversial. In 2019, surveillance was the biggest single end-use for AI in China, accounting for 53.8% of all AI-powered applications, according to research firm iiMedia. Finance was second with 15.8% followed by marketing (11.6%) and transport (4.2%). Some of the surveillance is alleged to have been directed “against Uygurs, ethnic Kazakhs, and other members of Muslim minority groups in the Xinjiang Uygur Autonomous Region[11],” and thus led to U.S. government blacklists of Sensetime, Megvii, and Yitu.

The emphasis on surveillance by China’s leading AI startups suggests there are big differences between China’s and America’s AI startup efforts. China is likely leading in surveillance while the U.S. is leading in other areas. By developing the ability to surveil their own citizens, China’s AI startups will develop the ability to export such systems to other authoritarian regimes. They will likely dominate the global surveillance market enabling authoritarian regimes to become stronger and more entrenched.

America’s (and other Western) AI startups will likely dominate other markets such as finance, news, advertising, logistics, and perhaps health care. News and advertising are already big users of AI[12]. When the other markets begin to grow is probably harder to say. Healthcare remains elusive with few successes.

Overall, I remain skeptical about AI. Although there are small success stories that will continue to proliferate, the media is obsessed with big fancy applications that are decades away, likely because they generate more page views than does the current reality. Many critics of DeepMind’s hype are also critical of the media’s excessive hype, a hype that I wrote about in my article for Issues in Science & Technology published one year ago, “Behind Technological Hype[13].”

That article also described the excessive hype of startups and technology by consultants and business schools, particularly entrepreneurship programs and strategy professors. Consultants and business schools must sell their entrepreneurship and innovation programs to prospective students and what better way to tell them that AI and other technologies will change the world, thus providing opportunities for students who study these issues. The number of books on how AI will change business and the strategies is enormous with few displaying an understanding of the challenges. Business schools love to say, it’s not the technology it’s the business model, and thus continue to ignore the key role the technology plays in determining the value proposition and customers for early AI applications[14].

The reality is that the $15 trillion in economic gains promised by some consulting companies by 2030 is likely overstated by at least 10 times and perhaps by 50 times. The best strategy is to find simple to implement applications or to serve customers with a high willingness to pay, a strategy I covered in a past Medium article[15].

What does this mean for applications in AI? A report, “The State of AI,” found that without major new research breakthroughs, dropping the ImageNet error rate from 11.5% to 1% would require over $100B! The report goes on to say that “We’re rapidly approaching outrageous computational, economic, and environmental costs to gain incrementally smaller improvements in model performance.” The good news is that the computing time to achieve the same accuracy has been decreasing by a factor of 2 every 16 months[16]. This means that startups should focus on applications that do not require high accuracies, another way of saying simple to implement applications. Of course, if you can find simple applications for which the customer has a high willingness to pay, profits will be even easier to obtain.

[1] https://www.asmag.com/showpost/26505.aspx#:~:text=According%20to%20the%20latest%20finance,loss%20of%20US%24621%20million.&text=For%20the%20past%20year%2C%20Nest,in%20the%20smart%20home%20market

[2] https://www.androidheadlines.com/2019/08/deepminds-revenue-dwarfed-by-losses-debt-to-alphabet.html

[3] https://www.businessinsider.com/google-deepmind-ai-startup-technology-2020-12

[4] https://kr-asia.com/how-is-megvii-doubling-down-listing-groundwork

[5] https://www.spglobal.com/marketintelligence/en/news-insights/latest-news-headlines/softbank-backed-cloudminds-us-ipo-chances-dwindle-57027223

[6] https://jeffreyleefunk.medium.com/small-impact-of-lockdowns-on-startup-unicorn-profits-only-7-of-69-are-profitable-in-2020-up-from-5a9b9d9b4c63

https://jeffreyleefunk.medium.com/can-ride-sharing-survive-huge-losses-and-lockdowns-d5bde44932d5

https://jeffreyleefunk.medium.com/only-6-of-73-unicorn-startups-are-profitable-and-none-did-recent-ipos-287d5c7ac8d0

[7] https://www.nature.com/articles/s41586-019-1799-6

[8] https://www.1843magazine.com/features/deepmind-and-google-the-battle-to-control-artificial-intelligence

[9] https://www.nature.com/articles/s41586-020-2766-y

[10] https://www.businessinsider.com/deepmind-google-protein-folding-ai-alphafold-technology-2020-12